The microdata is a sub-set of the ACS sample. As a subset, one would expect that it would not produce the same results as shown in the summary tables. However, I would at least expect it to not vary significantly and to be within the margins of error of the summary table. However, that is not what I am finding when I use the data.census.gov microdata tool. Is this an issue with the microdata tool since it is beta or something else causing the large discrepancies? For example, I was comparing the 2022 ACS 1-Year class of worker data for Virginia using summary table S2408 and a microdata query of the same data, and it shows significantly different numbers for each class of worker as you will see if you enter the following URL's.
S2408 Table: https://data.census.gov/table/ACSST1Y2022.S2408?q=class%20of%20worker%20Virginia
Microdata: https://data.census.gov/mdat/#/search?ds=ACSPUMS1Y2022&rv=COW&nv=ucgid&wt=PWGTP&g=0400000US51
I did not compare numbers. My only comment is on your microdata setup. I did not see you account for labor force status. The tabular data stipulates "employed". Your micro did not. Maybe try adding an employed labor force status element to your micro setup....?